Skip to content
SQL interview practice

SQL interview questions you can run.

A SQL interview for an analyst or analytics engineer role rarely asks for syntax. It gives you a business problem and a few tables, and checks whether you get the grain, the join, and the definition right. These 10 questions are written that way. Each one runs in your browser on linked tables, and your result is checked.

Free · No signup to run or check

What the interviewer is checking

Four things, in nearly every SQL screen.

Grain
What one row means in each table, and what it means after your join. Most wrong totals are a row counted twice.
Joins that keep the right rows
When a LEFT JOIN is needed, and where a filter goes so that it does not quietly turn into an inner join.
Window functions
Ranking inside a group, comparing a row with the one before it, and running totals. These come up in almost every live screen.
Definitions
Which rows belong in the denominator, and what counts as late, churned, or repeat. Say the definition before you write the query.
The questions

10 questions, from a warm-up to the ones that decide the round.

Each entry shows the question as you would hear it, the skill it tests, and the mistake that most often fails it. Try the question before you read the mistake.

  1. Double charges, or retries?

    Finance has noticed that some orders have more than one row in payments and is worried customers were charged twice.

    What it tests
    GROUP BY with HAVING, and reading the result before answering
    The mistake that fails it
    Reporting every order with two payment rows as a double charge
  2. Conversion rate without losing the leads that did not convert

    Marketing wants the share of leads that became purchases for each channel.

    What it tests
    LEFT JOIN, and where a filter on the joined table belongs
    The mistake that fails it
    Filtering conversion_type in WHERE, which removes every lead that never purchased
  3. Reconcile payments and refunds by order status

    The finance team wants to tie order status to money. For each order_status, show what was collected and what was returned.

    What it tests
    Aggregating to a shared grain before joining
    The mistake that fails it
    Joining orders to payments and refunds in one step, which repeats rows and inflates both totals
  4. Second-highest salary in each department

    The classic ranking question, on tables you can query. For each department, find the second-highest base salary among active employees.

    What it tests
    Window functions, and the difference between ROW_NUMBER, RANK, and DENSE_RANK
    The mistake that fails it
    Using ROW_NUMBER, which returns the top salary again when two people share it
  5. Top two products in each category

    The merchandising team wants the two best-selling products in every category.

    What it tests
    Top N per group with a window function over an aggregate
    The mistake that fails it
    ORDER BY revenue with LIMIT 2, which returns two products overall instead of two per category
  6. Running spend against budget

    A campaign manager wants to see how quickly campaign 'CMP0001' used its budget.

    What it tests
    Running totals with a window frame
    The mistake that fails it
    Leaving out ORDER BY in the window, which returns the campaign total on every row
  7. Which suppliers deliver late?

    Operations wants to know which suppliers deliver late most often.

    What it tests
    Choosing the denominator, and a minimum sample size
    The mistake that fails it
    Dividing by all purchase orders, including ones that have not arrived yet
  8. Revenue fell. Show the month-over-month change

    Leadership says revenue fell during 2024 and wants to see when.

    What it tests
    LAG, and the order in which a filter and a window function run
    The mistake that fails it
    Filtering to 2024 before calling LAG, which leaves January with nothing to compare against
  9. Monthly churn rate, with the right denominator

    The board wants a monthly churn rate for 2025. Last quarter's figure was challenged because nobody could say what it had been divided by.

    What it tests
    Building a month spine, and point-in-time counts
    The mistake that fails it
    Dividing cancellations by all subscriptions ever created instead of those active at the start of the month
  10. How long until a customer orders again?

    The retention team wants to know how long customers wait between orders, by customer segment.

    What it tests
    LAG partitioned by customer, and date arithmetic
    The mistake that fails it
    Averaging per customer and then averaging those averages, which gives a different number from the average gap
How to practice

Treat each one like the live screen.

  1. 01

    Say the definition first

    Before writing anything, state what one row of your result is and what the metric counts. In an interview this is where most of the marks are.

  2. 02

    Give yourself fifteen minutes

    Write and run the query without hints. Look at the rows it returns and ask whether the numbers are plausible before you check it.

  3. 03

    Explain what could go wrong

    Name the assumption your query depends on, such as one payment per order or no missing months. Then read the worked answer.

Questions

About this question set

What SQL is asked in an analytics engineer interview?
Mostly joins, aggregation, and window functions applied to a business question: reconciling two tables, ranking within a group, comparing a month with the one before it, and computing a rate with the right denominator. Interviewers pay more attention to whether you state the grain and the definition than to whether you remember a function name.
Which SQL dialect do these questions use?
They run on SQLite in your browser. The joins, CTEs, and window functions are standard SQL and work the same way in BigQuery, Snowflake, and Postgres. Date handling is the part that differs between engines: these answers use substr, date, and julianday where a warehouse would use DATE_TRUNC and DATEDIFF or DATE_DIFF.
Do I need an account to answer them?
No. Every question runs and is checked without an account. A free account adds hints, keeps track of what you have solved, and shows the worked answer once you have attempted a question.
Is the data real?
It is synthetic, generated to behave like business tables: orders with retried payments and partial refunds, subscriptions that cancel, purchase orders that arrive late. Each question runs on a linked sample of one of the free datasets on this site, and the full files can be downloaded for more practice.
How is an answer checked?
Your query runs in the browser and its first ten rows are compared with the expected result, in order. Column names do not have to match, but the columns must be in the order the question lists them and the rows must be sorted as asked.